Add Jefe::restart_me and use it for net watchdog.

This adds a method for tasks to restart themselves without indicating a
fault, which has the advantage of bypassing Jefe's crash-dump logic for
deliberate restarts.

Currently the only case of a deliberate restart is in net, for the
receive watchdog. This should knock about 66ms off each firing of that
watchdog.

I initially implemented this as a syscall, but @mkeeter noted it could
be done entirely outside the kernel -- and so here it is. Long live the
emokernel.

app/demo-stm32h7-nucleo/app-h743.toml

@@ -75,7 +75,7 @@ uses = ["eth", "eth_dma", "tim16"]
 start = true
 notifications = ["eth-irq", "mdio-timer-irq", "wake-timer"]
 interrupts = {"eth.irq" = "eth-irq", "tim16.irq" = "mdio-timer-irq"}
-task-slots = ["sys"]
+task-slots = ["sys", "jefe"]
 
 [tasks.user_leds]
 name = "drv-user-leds"

app/demo-stm32h7-nucleo/app-h753.toml

@@ -83,7 +83,7 @@ uses = ["eth", "eth_dma", "tim16"]
 start = true
 notifications = ["eth-irq", "mdio-timer-irq", "wake-timer"]
 interrupts = {"eth.irq" = "eth-irq", "tim16.irq" = "mdio-timer-irq"}
-task-slots = ["sys"]
+task-slots = ["sys", "jefe"]
 
 [tasks.user_leds]
 name = "drv-user-leds"

app/gimletlet/app-mgmt.toml

@@ -58,7 +58,7 @@ sections = {eth_bulk = "sram1"}
 uses = ["eth", "eth_dma", "tim16", "spi2"]
 start = true
 notifications = ["eth-irq", "mdio-timer-irq", "spi-irq", "wake-timer"]
-task-slots = ["sys", "user_leds"]
+task-slots = ["sys", "user_leds", "jefe"]
 
 [tasks.net.interrupts]
 "eth.irq" = "eth-irq"

app/gimletlet/app.toml

@@ -179,7 +179,7 @@ max-sizes = {flash = 131072, ram = 65536, sram1 = 16384}
 sections = {eth_bulk = "sram1"}
 uses = ["eth", "eth_dma", "tim16", "spi4"]
 start = true
-task-slots = ["sys"]
+task-slots = ["sys", "jefe"]
 notifications = ["eth-irq", "mdio-timer-irq", "spi-irq", "wake-timer"]
 
 [tasks.net.interrupts]

app/sidecar/rev-b.toml

@@ -72,7 +72,7 @@ sections = {eth_bulk = "sram1"}
 uses = ["eth", "eth_dma", "tim16", "spi3"]
 start = true
 notifications = ["eth-irq", "mdio-timer-irq", "spi-irq", "wake-timer"]
-task-slots = ["sys", "packrat", { seq = "sequencer" }]
+task-slots = ["sys", "packrat", { seq = "sequencer" }, "jefe"]
 
 [tasks.net.interrupts]
 "eth.irq" = "eth-irq"

app/sidecar/rev-c.toml

@@ -72,7 +72,7 @@ sections = {eth_bulk = "sram1"}
 uses = ["eth", "eth_dma", "tim16", "spi3"]
 start = true
 notifications = ["eth-irq", "mdio-timer-irq", "spi-irq", "wake-timer"]
-task-slots = ["sys", "packrat", { seq = "sequencer" }]
+task-slots = ["sys", "packrat", { seq = "sequencer" }, "jefe"]
 
 [tasks.net.interrupts]
 "eth.irq" = "eth-irq"

idl/jefe.idol

@@ -94,5 +94,17 @@ Interface(
             ),
             encoding: Hubpack,
         ),
+
+        // Note: this is the "raw" API; there is a nice wrapper in the client
+        // crate.
+        "restart_me_raw": (
+            description: "restarts the caller without recording a fault",
+            args: {},
+            // Note: this will not actually return, but Idol can't currently
+            // describe a noreturn IPC, so, we have a placeholder:
+            reply: Simple("()"),
+            encoding: Hubpack,
+            idempotent: true,
+        ),
     },
 )

task/jefe-api/src/lib.rs

@@ -34,4 +34,13 @@ pub enum DumpAreaError {
     AlreadyInUse,
 }
 
+impl Jefe {
+    /// Asks the supervisor to restart the current task without recording a
+    /// fault.
+    pub fn restart_me(&self) -> ! {
+        self.restart_me_raw();
+        unreachable!()
+    }
+}
+
 include!(concat!(env!("OUT_DIR"), "/client_stub.rs"));

task/jefe/src/main.rs

@@ -207,6 +207,19 @@ impl idl::InOrderJefeImpl for ServerImpl<'_> {
         Ok(())
     }
 
+    fn restart_me_raw(
+        &mut self,
+        msg: &userlib::RecvMessage,
+    ) -> Result<(), RequestError<Infallible>> {
+        kipc::restart_task(msg.sender.index(), true);
+
+        // Note: the returned value here won't go anywhere because we just
+        // unblocked the caller. So this is doing a small amount of unnecessary
+        // work. This is a compromise because Idol can't easily describe an IPC
+        // that won't return at this time.
+        Ok(())
+    }
+
     cfg_if::cfg_if! {
         if #[cfg(feature = "dump")] {
             fn get_dump_area(

task/net/src/bsp/gimlet_bcd.rs

@@ -17,11 +17,9 @@ use task_jefe_api::Jefe;
 use task_net_api::{
     ManagementCounters, ManagementLinkStatus, MgmtError, PhyError,
 };
-use userlib::{sys_recv_closed, task_slot, FromPrimitive, TaskId};
+use userlib::{sys_recv_closed, FromPrimitive, TaskId};
 use vsc7448_pac::types::PhyRegisterAddress;
 
-task_slot!(JEFE, jefe);
-
 ////////////////////////////////////////////////////////////////////////////////
 
 pub struct BspImpl(mgmt::Bsp);
@@ -56,7 +54,7 @@ impl crate::bsp_support::Bsp for BspImpl {
     fn preinit() {
         // Wait for the sequencer to turn on the clock. This requires that Jefe
         // state change notifications are routed to our notification bit 3.
-        let jefe = Jefe::from(JEFE.get_task_id());
+        let jefe = Jefe::from(crate::JEFE.get_task_id());
 
         loop {
             // This laborious list is intended to ensure that new power states

task/net/src/bsp/psc_a.rs

@@ -17,11 +17,9 @@ use task_jefe_api::Jefe;
 use task_net_api::{
     ManagementCounters, ManagementLinkStatus, MgmtError, PhyError,
 };
-use userlib::{sys_recv_closed, task_slot, FromPrimitive, TaskId};
+use userlib::{sys_recv_closed, FromPrimitive, TaskId};
 use vsc7448_pac::types::PhyRegisterAddress;
 
-task_slot!(JEFE, jefe);
-
 ////////////////////////////////////////////////////////////////////////////////
 
 pub struct BspImpl(mgmt::Bsp);
@@ -55,7 +53,7 @@ impl bsp_support::Bsp for BspImpl {
 
     fn preinit() {
         // Wait for the sequencer to turn read our VPD.
-        let jefe = Jefe::from(JEFE.get_task_id());
+        let jefe = Jefe::from(crate::JEFE.get_task_id());
 
         loop {
             // This laborious list is intended to ensure that new power states

task/net/src/bsp/psc_bc.rs

@@ -17,11 +17,9 @@ use task_jefe_api::Jefe;
 use task_net_api::{
     ManagementCounters, ManagementLinkStatus, MgmtError, PhyError,
 };
-use userlib::{sys_recv_closed, task_slot, FromPrimitive, TaskId};
+use userlib::{sys_recv_closed, FromPrimitive, TaskId};
 use vsc7448_pac::types::PhyRegisterAddress;
 
-task_slot!(JEFE, jefe);
-
 ////////////////////////////////////////////////////////////////////////////////
 
 pub struct BspImpl(mgmt::Bsp);
@@ -60,7 +58,7 @@ impl bsp_support::Bsp for BspImpl {
 
     fn preinit() {
         // Wait for the sequencer to turn read our VPD.
-        let jefe = Jefe::from(JEFE.get_task_id());
+        let jefe = Jefe::from(crate::JEFE.get_task_id());
 
         loop {
             // This laborious list is intended to ensure that new power states

task/net/src/main.rs

@@ -59,6 +59,7 @@ use core::sync::atomic::{AtomicU32, Ordering};
 use enum_map::Enum;
 use multitimer::{Multitimer, Repeat};
 use task_net_api::MacAddressBlock;
+use task_jefe_api::Jefe;
 use zerocopy::{AsBytes, U16};
 
 #[cfg(feature = "h743")]
@@ -74,6 +75,7 @@ use crate::bsp::BspImpl;
 use crate::bsp_support::Bsp;
 
 task_slot!(SYS, sys);
+task_slot!(JEFE, jefe);
 
 #[cfg(feature = "vpd-mac")]
 task_slot!(PACKRAT, packrat);
@@ -153,6 +155,7 @@ static ITER_COUNT: AtomicU32 = AtomicU32::new(0);
 fn main() -> ! {
     let sys = SYS.get_task_id();
     let sys = Sys::from(sys);
+    let jefe = Jefe::from(JEFE.get_task_id());
 
     // Do any preinit tasks specific to this board.  For hardware which requires
     // explicit clock configuration, this is where the `net` tasks waits for
@@ -265,7 +268,9 @@ fn main() -> ! {
                         server.wake();
                         // timer is set to auto-repeat
                     }
-                    Timers::Watchdog => panic!("MAC RX watchdog"),
+                    Timers::Watchdog => {
+                        jefe.restart_me();
+                    }
                 }
             }
             let mut msgbuf = [0u8; idl::INCOMING_SIZE];